Energy Use Intensity

Data Science side project

All rights reserved by Thaiprayoon N.

Table of Contents

• Import data
  • Check values in each columns
• EDA and Data preparation.
  • Cut off the outlier
    • Plot some column to explore outlier
    • IQR (Interquartile Range)
  • Min-Max Normalization
    • Heat map plot to show the data correlation
• Predict by using XGboost
  • Show some of the predicted data
  • Mean Absolute Error
  • Show the first predicted data and MAE
  • 3D Scatter Plot of Predict datas Vs Real data Vs Error
  • Line Plot of Predict datas, Real data, and Error with 100 observations

Import data

This data set comes from Kaggle.com There are Building simulation results from for than 250k EnergyPlus simulations. With 48 columns of important data for using analytics energy consumption of building comsumtion following.

• Building Type
• Floor to floor height
• Number of floors (Total Area / Floor Area)
• Height (Number of floors * Floor to floor height)
• Wall R value
• Roof R value
• Window U value
• etc.

Universal Design Space Exploration

Design Space Exploration (DSE) analysis techniques represent a data-centric approach to integrating performance analysis in early design phases when there is the greatest potential to cheaply improve the energy efficiency of a building. We focus on a novel extension of DSE called Universal Design Space Exploration (UDSE), which leverages massive databases of pre-simulated analysis that represent all possible outcomes of common analysis workflows. These databases, called Design Spaces, become “universal” when a single pre-simulated design space can be re-applied to future, unknown projects. Unlike current simulation methods, which require a design to exist before it can be analyzed and often take minutes or hours to simulate, UDSE leverages pre-simulation to deliver rapid and relevant insight as new designs are conceptualized. The data underpinning UDSE enables advanced statistical and Artificial Intelligence methods, allowing UDSE to deliver a greater understanding of the larger problem being explored rather than simply delivering analysis of several pre-conceived design options. We believe that UDSE can provide instantaneous, relevant analysis for all building design projects at negligible cost.

AutoBEM

Oak Ridge National Laboratory has developed a collection of software and algorithms, collectively referred to as “Automatic Building Energy Modeling” (AutoBEM), which allows building energy modeling of each building at large geographic scales (AutoBEM). Within AutoBEM, building properties are detected, inferred, or predicted as inputs to generate building energy models using OpenStudio and simulate these buildings using EnergyPlus. OpenStudio is a collection of software tools to support energy modeling in EnergyPlus, which is a physical building energy simulation engine (OpenStudio) (EnergyPlus).

See more on Kaggle

import pandas as pd

data0 = pd.read_csv("Universal_Design_Space_Building_Energy_Simulation_input_output.csv",low_memory=False)
data0.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 260748 entries, 0 to 260747
Data columns (total 49 columns):
 #   Column                                 Non-Null Count   Dtype  
---  ------                                 --------------   -----  
 0   ID                                     260748 non-null  object 
 1   BuildingType                           260748 non-null  object 
 2   ClimateZone                            260748 non-null  object 
 3   TotalArea                              260748 non-null  int64  
 4   TotalArea_Setting                      260748 non-null  object 
 5   FloorArea                              260748 non-null  int64  
 6   FloorArea_Setting                      260748 non-null  object 
 7   NumFloors                              260748 non-null  int64  
 8   PlateDepth                             260748 non-null  int64  
 9   PlateDepth_Setting                     260748 non-null  object 
 10  PlateLength                            260748 non-null  int64  
 11  FloorHeight                            260748 non-null  int64  
 12  FloorHeight_Setting                    260748 non-null  object 
 13  Height                                 260748 non-null  int64  
 14  WWR                                    260748 non-null  float64
 15  WWR_surfaces                           260748 non-null  object 
 16  SolarDesign                            260748 non-null  object 
 17  Standard                               260748 non-null  object 
 18  HVAC                                   260748 non-null  object 
 19  HVAC_Setting                           260748 non-null  object 
 20  EnvelopeQuality_Setting                260748 non-null  object 
 21  Wall_R_Value                           260748 non-null  float64
 22  Roof_R_Value                           260748 non-null  float64
 23  Glass_and_Frame_U_Value                260748 non-null  float64
 24  SHGC                                   260748 non-null  float64
 25  LPD_Adjustment                         260748 non-null  float64
 26  LPD_Adjustment_Setting                 260748 non-null  object 
 27  Interior_Lights_Final_W_per_sf         260748 non-null  float64
 28  Exterior_Lights_Final_1_W              260748 non-null  float64
 29  Exterior_Lights_Final_2_W              260748 non-null  float64
 30  Setpoint_Setting                       260748 non-null  object 
 31  HeatingCoil                            260748 non-null  object 
 32  COP_Efficiency_Heating                 260748 non-null  object 
 33  CoolingCoil                            260748 non-null  object 
 34  COP_Efficiency_Cooling                 260748 non-null  float64
 35  EUI_kBTU_per_sf                        260748 non-null  float64
 36  Electricity_Facility_kBTU_per_sf       260748 non-null  float64
 37  NaturalGas_Facility_kBTU_per_sf        260748 non-null  float64
 38  Cooling_Electricity_kBTU_per_sf        260748 non-null  float64
 39  Heating_Electricity_kBTU_per_sf        260748 non-null  float64
 40  Heating_NaturalGas_kBTU_per_sf         260748 non-null  float64
 41  Heating_Total_kBTU_per_sf              260748 non-null  float64
 42  WaterSystems_Electricity_kBTU_per_sf   260748 non-null  float64
 43  Lighting_Electricity_kBTU_per_sf       260748 non-null  float64
 44  Equipment_Electricity_kBTU_per_sf      260748 non-null  float64
 45  Fans_Electricity_kBTU_per_sf           260748 non-null  float64
 46  Pumps_Electricity_kBTU_per_sf          260748 non-null  float64
 47  HeatRejection_Electricity_kBTU_per_sf  260748 non-null  float64
 48  HeatRecovery_Electricity_kBTU_per_sf   260748 non-null  float64
dtypes: float64(24), int64(7), object(18)
memory usage: 97.5+ MB

data = data0.copy()
data.describe()

	TotalArea	FloorArea	NumFloors	PlateDepth	PlateLength	FloorHeight	Height	WWR	Wall_R_Value	Roof_R_Value	...	Heating_Electricity_kBTU_per_sf	Heating_NaturalGas_kBTU_per_sf	Heating_Total_kBTU_per_sf	WaterSystems_Electricity_kBTU_per_sf	Lighting_Electricity_kBTU_per_sf	Equipment_Electricity_kBTU_per_sf	Fans_Electricity_kBTU_per_sf	Pumps_Electricity_kBTU_per_sf	HeatRejection_Electricity_kBTU_per_sf	HeatRecovery_Electricity_kBTU_per_sf
count	2.607480e+05	260748.000000	260748.000000	260748.000000	260748.000000	260748.000000	260748.000000	260748.000000	260748.000000	260748.000000	...	260748.000000	260748.000000	260748.000000	260748.0	260748.000000	260748.000000	260748.000000	260748.000000	260748.000000	260748.000000
mean	1.994595e+05	25846.604507	8.866783	97.235906	209.707273	14.602774	129.436559	59.377107	22.190043	41.705422	...	0.537301	43.137857	43.675158	0.0	8.084788	38.817513	22.115252	2.509235	0.395789	4.833672
std	2.156952e+05	10350.688871	10.753368	25.367408	76.845830	1.970821	161.182146	17.593025	7.072464	14.076231	...	1.728176	97.983243	97.762052	0.0	4.609707	15.903090	32.499733	4.598613	1.391958	10.479290
min	3.994400e+04	14519.000000	1.000000	45.000000	129.000000	10.000000	13.000000	25.000000	8.330000	21.340000	...	0.000000	0.000000	0.000000	0.0	0.648153	11.218701	0.180277	0.000000	0.000000	0.000000
25%	1.016330e+05	14971.000000	4.000000	75.000000	140.000000	13.000000	52.000000	50.390625	16.330000	31.340000	...	0.000000	0.000000	0.232082	0.0	4.537323	24.806507	2.727917	0.016485	0.000000	0.000000
50%	1.497120e+05	24998.000000	6.000000	98.000000	197.000000	15.000000	85.000000	61.742424	20.330000	31.340000	...	0.000000	0.688570	1.768745	0.0	7.506889	41.608445	7.107739	1.186714	0.000000	0.084560
75%	2.018510e+05	39539.000000	10.000000	122.000000	238.000000	16.000000	135.000000	71.228070	30.000000	60.000000	...	0.071649	8.487569	9.333217	0.0	10.513514	52.364630	15.606273	2.963894	0.229866	1.419913
max	1.009256e+06	40434.000000	67.000000	153.000000	404.000000	18.000000	1072.000000	89.910314	30.000000	60.000000	...	39.978431	583.250051	583.250051	0.0	19.497817	68.665014	103.441310	34.478568	15.186499	33.107724

8 rows × 31 columns

data.head()

	ID	BuildingType	ClimateZone	TotalArea	TotalArea_Setting	FloorArea	FloorArea_Setting	NumFloors	PlateDepth	PlateDepth_Setting	...	Heating_Electricity_kBTU_per_sf	Heating_NaturalGas_kBTU_per_sf	Heating_Total_kBTU_per_sf	WaterSystems_Electricity_kBTU_per_sf	Lighting_Electricity_kBTU_per_sf	Equipment_Electricity_kBTU_per_sf	Fans_Electricity_kBTU_per_sf	Pumps_Electricity_kBTU_per_sf	HeatRejection_Electricity_kBTU_per_sf	HeatRecovery_Electricity_kBTU_per_sf
0	College_1A_100000_14286_120_13_25_20_25_BaseLi...	College	1A	101793	low	14542	low	7	122	high	...	0.083671	0.0	0.083671	0.0	10.209886	44.423582	5.378116	4.748715	0.0	3.704229
1	College_1A_100000_14286_120_13_25_20_25_BaseLi...	College	1A	101793	low	14542	low	7	122	high	...	0.083671	0.0	0.083671	0.0	10.209886	44.423582	3.689256	4.675590	0.0	3.647120
2	College_1A_100000_14286_120_13_25_20_25_BaseLi...	College	1A	101793	low	14542	low	7	122	high	...	0.083856	0.0	0.083856	0.0	10.209886	44.423582	5.725152	4.350329	0.0	3.657521
3	College_1A_100000_14286_120_13_25_20_25_BaseLi...	College	1A	101793	low	14542	low	7	122	high	...	0.083856	0.0	0.083856	0.0	10.209886	44.423582	3.751486	4.224587	0.0	3.551803
4	College_1A_100000_14286_120_13_25_20_25_BaseLi...	College	1A	101793	low	14542	low	7	122	high	...	0.062975	0.0	0.062975	0.0	10.209886	44.423582	4.730901	4.131386	0.0	3.586181

5 rows × 49 columns

Check values in each columns

data.columns

Index(['ID', 'BuildingType', 'ClimateZone', 'TotalArea', 'TotalArea_Setting',
       'FloorArea', 'FloorArea_Setting', 'NumFloors', 'PlateDepth',
       'PlateDepth_Setting', 'PlateLength', 'FloorHeight',
       'FloorHeight_Setting', 'Height', 'WWR', 'WWR_surfaces', 'SolarDesign',
       'Standard', 'HVAC', 'HVAC_Setting', 'EnvelopeQuality_Setting',
       'Wall_R_Value', 'Roof_R_Value', 'Glass_and_Frame_U_Value', 'SHGC',
       'LPD_Adjustment', 'LPD_Adjustment_Setting',
       'Interior_Lights_Final_W_per_sf', 'Exterior_Lights_Final_1_W',
       'Exterior_Lights_Final_2_W', 'Setpoint_Setting', 'HeatingCoil',
       'COP_Efficiency_Heating', 'CoolingCoil', 'COP_Efficiency_Cooling',
       'EUI_kBTU_per_sf', 'Electricity_Facility_kBTU_per_sf',
       'NaturalGas_Facility_kBTU_per_sf', 'Cooling_Electricity_kBTU_per_sf',
       'Heating_Electricity_kBTU_per_sf', 'Heating_NaturalGas_kBTU_per_sf',
       'Heating_Total_kBTU_per_sf', 'WaterSystems_Electricity_kBTU_per_sf',
       'Lighting_Electricity_kBTU_per_sf', 'Equipment_Electricity_kBTU_per_sf',
       'Fans_Electricity_kBTU_per_sf', 'Pumps_Electricity_kBTU_per_sf',
       'HeatRejection_Electricity_kBTU_per_sf',
       'HeatRecovery_Electricity_kBTU_per_sf'],
      dtype='object')

# Check unique values
# col = [i for i in data.columns]
# for i in col:
#     print(i,"---> \n",data[i].unique(),"\n")

data = data.drop(columns=["TotalArea_Setting","FloorArea_Setting","PlateDepth_Setting",
                          "FloorHeight_Setting","WWR_surfaces","Standard","HVAC",
                          "WaterSystems_Electricity_kBTU_per_sf","LPD_Adjustment",
                          "Setpoint_Setting"])

EDA and Data preparation.

EDA: Exploratory Data Analysis

Cut off the outlier

# Changing Dtype object to category to numeric

df2 = data.copy()
g = df2.columns.to_series().groupby(df2.dtypes).groups
a = {k.name: v for k, v in g.items()}
obls = []
flols = []
intls = []
for i in a['object']:
    obls.append(i)
for i in a['float64']:
    flols.append(i)
for i in a['int64']:
    intls.append(i)

Plot some column to explore outlier

import matplotlib.pyplot as plt
plt.figure(figsize=(15,5))

data.boxplot()
plt.xticks(rotation = 90)

(array([ 1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16, 17,
        18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29]),
 [Text(1, 0, 'TotalArea'),
  Text(2, 0, 'FloorArea'),
  Text(3, 0, 'NumFloors'),
  Text(4, 0, 'PlateDepth'),
  Text(5, 0, 'PlateLength'),
  Text(6, 0, 'FloorHeight'),
  Text(7, 0, 'Height'),
  Text(8, 0, 'WWR'),
  Text(9, 0, 'Wall_R_Value'),
  Text(10, 0, 'Roof_R_Value'),
  Text(11, 0, 'Glass_and_Frame_U_Value'),
  Text(12, 0, 'SHGC'),
  Text(13, 0, 'Interior_Lights_Final_W_per_sf'),
  Text(14, 0, 'Exterior_Lights_Final_1_W'),
  Text(15, 0, 'Exterior_Lights_Final_2_W'),
  Text(16, 0, 'COP_Efficiency_Cooling'),
  Text(17, 0, 'EUI_kBTU_per_sf'),
  Text(18, 0, 'Electricity_Facility_kBTU_per_sf'),
  Text(19, 0, 'NaturalGas_Facility_kBTU_per_sf'),
  Text(20, 0, 'Cooling_Electricity_kBTU_per_sf'),
  Text(21, 0, 'Heating_Electricity_kBTU_per_sf'),
  Text(22, 0, 'Heating_NaturalGas_kBTU_per_sf'),
  Text(23, 0, 'Heating_Total_kBTU_per_sf'),
  Text(24, 0, 'Lighting_Electricity_kBTU_per_sf'),
  Text(25, 0, 'Equipment_Electricity_kBTU_per_sf'),
  Text(26, 0, 'Fans_Electricity_kBTU_per_sf'),
  Text(27, 0, 'Pumps_Electricity_kBTU_per_sf'),
  Text(28, 0, 'HeatRejection_Electricity_kBTU_per_sf'),
  Text(29, 0, 'HeatRecovery_Electricity_kBTU_per_sf')])

plt.figure(figsize=(5,8))
data[['Exterior_Lights_Final_1_W','Exterior_Lights_Final_2_W']].boxplot()
plt.xticks(rotation = 45)

(array([1, 2]),
 [Text(1, 0, 'Exterior_Lights_Final_1_W'),
  Text(2, 0, 'Exterior_Lights_Final_2_W')])

import seaborn as sns
import matplotlib.pyplot as plt

for i in flols[5:10]:
    fig, ax = plt.subplots(1, 2,figsize=(15,5))

    ax[0].set_title(i)
    ax[1].set_title('Distribution plot')
    plt.show
    sns.boxplot(ax=ax[0], data=data[i], orient='h')
    sns.histplot(ax=ax[1], data=data[i])

IQR (Interquartile Range)

$$Interquartile\ Range = Q_{3} - Q_{1}$$

Cut off outlier data that is below $3^{rd}$quartile ($25^{th}$percentile) and above $3^{rd}$quartile ($75^{th}$percentile). \ Ref: Interquartile Range

# import seaborn as sns
# import matplotlib.pyplot as plt

# for i in flols:
#     fig, ax = plt.subplots(1, 2,figsize=(15,5))

#     ax[0].set_title('Boxplot')
#     ax[1].set_title('Distribution plot')

#     sns.boxplot(ax=ax[0], data=data[i], orient='h')
#     sns.histplot(ax=ax[1], data=data[i])

dfiqr = data.copy()

for i in intls:
    dfiqr = dfiqr.drop(columns=[i])
for j in obls:
    dfiqr = dfiqr.drop(columns=[j])

#q1,q3 =np.percentile(data,[25,75], axis=0)
tmp = dfiqr.quantile(([0.25,0.75]))
# print(tmp)
q1 = tmp.iloc[0,:]
q3 = tmp.iloc[1,:]

IQR = q3-q1
# print("IQR\n",IQR)
upper = q3+1.5*IQR
lower = q1-1.5*IQR
# print("The largest value in the data set \n",upper)
# print("The smallest value in the data set \n",lower)

# Find index to drop
result = dfiqr[((dfiqr>upper).any(axis = 1)) | ((dfiqr<lower).any(axis = 1))].index
result

Int64Index([     0,      1,      2,      3,      4,      5,      6,      7,
                 8,      9,
            ...
            260738, 260739, 260740, 260741, 260742, 260743, 260744, 260745,
            260746, 260747],
           dtype='int64', length=158695)

dfiqr["ID"] = data["ID"]
dfiqrnew = dfiqr.drop(index=result)

dfiqrnew.head()

	WWR	Wall_R_Value	Roof_R_Value	Glass_and_Frame_U_Value	SHGC	Interior_Lights_Final_W_per_sf	Exterior_Lights_Final_1_W	Exterior_Lights_Final_2_W	COP_Efficiency_Cooling	EUI_kBTU_per_sf	...	Heating_Electricity_kBTU_per_sf	Heating_NaturalGas_kBTU_per_sf	Heating_Total_kBTU_per_sf	Lighting_Electricity_kBTU_per_sf	Equipment_Electricity_kBTU_per_sf	Fans_Electricity_kBTU_per_sf	Pumps_Electricity_kBTU_per_sf	HeatRejection_Electricity_kBTU_per_sf	HeatRecovery_Electricity_kBTU_per_sf	ID
11	25.000000	30.00	60.00	0.20	0.25	1.0	441.80	8552.0	3.552525	94.998433	...	0.050334	0.0	0.050334	10.209886	44.423582	3.296442	3.845008	0.0	3.337595	College_1A_100000_14286_120_13_25_20_25_BaseLi...
17	25.000000	30.00	60.00	0.20	0.25	1.0	441.80	8552.0	3.996591	89.394823	...	0.040897	0.0	0.040897	10.209886	44.423582	3.296442	3.845008	0.0	3.337595	College_1A_100000_14286_120_13_25_20_25_BaseLi...
23	25.000000	30.00	60.00	0.20	0.25	1.0	441.80	8552.0	4.440657	83.791214	...	0.031459	0.0	0.031459	10.209886	44.423582	3.296442	3.845008	0.0	3.337595	College_1A_100000_14286_120_13_25_20_25_BaseLi...
29	25.000000	8.33	21.34	0.66	0.25	0.4	176.72	3420.8	2.960438	94.982676	...	0.135453	0.0	0.135453	4.083955	44.423582	3.185673	3.685125	0.0	3.213624	College_1A_100000_14286_120_13_25_66_25_BestLi...
31	74.152542	8.33	21.34	0.66	0.25	0.4	176.72	3420.8	3.552525	93.761766	...	0.172216	0.0	0.172216	4.083955	44.423582	3.352672	4.427529	0.0	3.426734	College_1A_100000_14286_120_13_25_66_25_BestLi...

5 rows × 23 columns

Min-Max Normalization

dfiqrnew2 = dfiqrnew.copy()
for i in flols:
    mu = dfiqrnew[i].mean()
    sd = dfiqrnew[i].std()
#     print(mu, sd)

    max = dfiqrnew[i].max()
    min = dfiqrnew[i].min()
#     print(max, min)

#     dfiqrnew[i] = dfiqrnew[i].apply(lambda x:(x-mu)/sd)
    dfiqrnew2[i] = dfiqrnew[i].apply(lambda x:(x-min)/(max-min))
#df['size'] = df['size'].apply(lambda x:(x-mu)/sd if sd!=0 else x)
dfiqrnew2.head()

	WWR	Wall_R_Value	Roof_R_Value	Glass_and_Frame_U_Value	SHGC	Interior_Lights_Final_W_per_sf	Exterior_Lights_Final_1_W	Exterior_Lights_Final_2_W	COP_Efficiency_Cooling	EUI_kBTU_per_sf	...	Heating_Electricity_kBTU_per_sf	Heating_NaturalGas_kBTU_per_sf	Heating_Total_kBTU_per_sf	Lighting_Electricity_kBTU_per_sf	Equipment_Electricity_kBTU_per_sf	Fans_Electricity_kBTU_per_sf	Pumps_Electricity_kBTU_per_sf	HeatRejection_Electricity_kBTU_per_sf	HeatRecovery_Electricity_kBTU_per_sf	ID
11	0.000000	1.0	1.0	0.0	0.0	1.000000	0.034954	1.0	0.091653	0.582000	...	0.281023	0.0	0.002372	0.646387	0.578016	0.086309	0.723544	0.0	0.940548	College_1A_100000_14286_120_13_25_20_25_BaseLi...
17	0.000000	1.0	1.0	0.0	0.0	1.000000	0.034954	1.0	0.160393	0.537886	...	0.228331	0.0	0.001927	0.646387	0.578016	0.086309	0.723544	0.0	0.940548	College_1A_100000_14286_120_13_25_20_25_BaseLi...
23	0.000000	1.0	1.0	0.0	0.0	1.000000	0.034954	1.0	0.229133	0.493772	...	0.175639	0.0	0.001483	0.646387	0.578016	0.086309	0.723544	0.0	0.940548	College_1A_100000_14286_120_13_25_20_25_BaseLi...
29	0.000000	0.0	0.0	1.0	0.0	0.294118	0.010065	0.4	0.000000	0.581876	...	0.756249	0.0	0.006384	0.232265	0.578016	0.083109	0.693458	0.0	0.905612	College_1A_100000_14286_120_13_25_66_25_BestLi...
31	0.757238	0.0	0.0	1.0	0.0	0.294118	0.010065	0.4	0.091653	0.572265	...	0.961502	0.0	0.008116	0.232265	0.578016	0.087934	0.833161	0.0	0.965667	College_1A_100000_14286_120_13_25_66_25_BestLi...

5 rows × 23 columns

data2 = data.copy()

for i in flols:
    data2 = data2.drop(columns=[i])
    
dfn = pd.merge(data2,dfiqrnew2, on = "ID",how = "inner")
dfn.head()

	ID	BuildingType	ClimateZone	TotalArea	FloorArea	NumFloors	PlateDepth	PlateLength	FloorHeight	Height	...	Cooling_Electricity_kBTU_per_sf	Heating_Electricity_kBTU_per_sf	Heating_NaturalGas_kBTU_per_sf	Heating_Total_kBTU_per_sf	Lighting_Electricity_kBTU_per_sf	Equipment_Electricity_kBTU_per_sf	Fans_Electricity_kBTU_per_sf	Pumps_Electricity_kBTU_per_sf	HeatRejection_Electricity_kBTU_per_sf	HeatRecovery_Electricity_kBTU_per_sf
0	College_1A_100000_14286_120_13_25_20_25_BaseLi...	College	1A	101793	14542	7	122	350	13	91	...	0.821321	0.281023	0.0	0.002372	0.646387	0.578016	0.086309	0.723544	0.0	0.940548
1	College_1A_100000_14286_120_13_25_20_25_BaseLi...	College	1A	101793	14542	7	122	350	13	91	...	0.667235	0.228331	0.0	0.001927	0.646387	0.578016	0.086309	0.723544	0.0	0.940548
2	College_1A_100000_14286_120_13_25_20_25_BaseLi...	College	1A	101793	14542	7	122	350	13	91	...	0.513149	0.175639	0.0	0.001483	0.646387	0.578016	0.086309	0.723544	0.0	0.940548
3	College_1A_100000_14286_120_13_25_66_25_BestLi...	College	1A	101793	14542	7	122	350	13	91	...	0.998144	0.756249	0.0	0.006384	0.232265	0.578016	0.083109	0.693458	0.0	0.905612
4	College_1A_100000_14286_120_13_25_66_25_BestLi...	College	1A	101793	14542	7	122	350	13	91	...	0.932584	0.961502	0.0	0.008116	0.232265	0.578016	0.087934	0.833161	0.0	0.965667

5 rows × 39 columns

# Convert Dtype object to category to numeric
df2 = dfn.copy()

for i in obls:
    df2[i] = pd.Categorical(df2[i])
    df2[i] = df2[i].cat.codes
# df2.info()

# Convert to float
from sklearn.preprocessing import LabelEncoder

col = df2.columns
le = LabelEncoder()
for i in col:
    df2[i] = df2[i].astype('float')
df2.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 102053 entries, 0 to 102052
Data columns (total 39 columns):
 #   Column                                 Non-Null Count   Dtype  
---  ------                                 --------------   -----  
 0   ID                                     102053 non-null  float64
 1   BuildingType                           102053 non-null  float64
 2   ClimateZone                            102053 non-null  float64
 3   TotalArea                              102053 non-null  float64
 4   FloorArea                              102053 non-null  float64
 5   NumFloors                              102053 non-null  float64
 6   PlateDepth                             102053 non-null  float64
 7   PlateLength                            102053 non-null  float64
 8   FloorHeight                            102053 non-null  float64
 9   Height                                 102053 non-null  float64
 10  SolarDesign                            102053 non-null  float64
 11  HVAC_Setting                           102053 non-null  float64
 12  EnvelopeQuality_Setting                102053 non-null  float64
 13  LPD_Adjustment_Setting                 102053 non-null  float64
 14  HeatingCoil                            102053 non-null  float64
 15  COP_Efficiency_Heating                 102053 non-null  float64
 16  CoolingCoil                            102053 non-null  float64
 17  WWR                                    102053 non-null  float64
 18  Wall_R_Value                           102053 non-null  float64
 19  Roof_R_Value                           102053 non-null  float64
 20  Glass_and_Frame_U_Value                102053 non-null  float64
 21  SHGC                                   102053 non-null  float64
 22  Interior_Lights_Final_W_per_sf         102053 non-null  float64
 23  Exterior_Lights_Final_1_W              102053 non-null  float64
 24  Exterior_Lights_Final_2_W              102053 non-null  float64
 25  COP_Efficiency_Cooling                 102053 non-null  float64
 26  EUI_kBTU_per_sf                        102053 non-null  float64
 27  Electricity_Facility_kBTU_per_sf       102053 non-null  float64
 28  NaturalGas_Facility_kBTU_per_sf        102053 non-null  float64
 29  Cooling_Electricity_kBTU_per_sf        102053 non-null  float64
 30  Heating_Electricity_kBTU_per_sf        102053 non-null  float64
 31  Heating_NaturalGas_kBTU_per_sf         102053 non-null  float64
 32  Heating_Total_kBTU_per_sf              102053 non-null  float64
 33  Lighting_Electricity_kBTU_per_sf       102053 non-null  float64
 34  Equipment_Electricity_kBTU_per_sf      102053 non-null  float64
 35  Fans_Electricity_kBTU_per_sf           102053 non-null  float64
 36  Pumps_Electricity_kBTU_per_sf          102053 non-null  float64
 37  HeatRejection_Electricity_kBTU_per_sf  102053 non-null  float64
 38  HeatRecovery_Electricity_kBTU_per_sf   102053 non-null  float64
dtypes: float64(39)
memory usage: 31.1 MB

Heat map plot to show the data correlation

import matplotlib.pyplot as plt
import seaborn as sns
plt.figure(figsize=(24, 15))


data2 = df2.copy()
data2 = data2.iloc[:,1:27]
col = data2.columns
corr_sp = data2[col].corr(method='pearson')
# plot correlation
sns.heatmap(abs(corr_sp), annot=True, cmap="flare") # summer_r
plt.title('Correlation Heatmap', fontdict={'fontsize':24}, pad=12)
plt.show()

Predict by using XGboost

import xgboost as xgb
from sklearn.metrics import mean_absolute_error
from sklearn.model_selection import train_test_split
import pandas as pd
import numpy as np

dfnew = df2.copy()
X, y = dfnew.iloc[:,1:26],dfnew.iloc[:,-13]
data_dmatrix = xgb.DMatrix(data=X,label=y)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=999)

xg_reg = xgb.XGBRegressor(objective ='reg:squarederror', colsample_bytree = 0.3, learning_rate = 0.1,
                max_depth = 5, alpha = 10, n_estimators = 10)

xg_reg.fit(X_train,y_train)

preds = xg_reg.predict(X_test)


# reverse score
# mu = dfiqrnew['EUI_kBTU_per_sf'].mean()

max1 = dfiqrnew['EUI_kBTU_per_sf'].max()
min1 = dfiqrnew['EUI_kBTU_per_sf'].min()

y_test_new = (y_test*(max1-min1)) + min1
preds_new = (preds*(max1-min1)) + min1

print("Mean Absolute Error: ",mean_absolute_error(y_test_new, preds_new))

Mean Absolute Error:  9.72474943313263

Show some of the predicted data

X_test2 = X_test.copy()
X_test2['PREDICT'] = preds_new
X_test2['REAL DATA'] = data['EUI_kBTU_per_sf']
X_test2['ERROR'] = abs(X_test2['PREDICT'] - X_test2['REAL DATA'])

X_test2.iloc[:5,-3:]

	PREDICT	REAL DATA	ERROR
36048	51.335945	68.263557	16.927612
94705	89.275932	70.950452	18.325480
21615	76.678230	64.576512	12.101718
1480	80.930763	84.433132	3.502369
92790	87.536217	59.044387	28.491830

Mean Absolute Error

MAE: Mean Absolute Error

Source: wikipedia.org

Show the first predicted data and MAE

print("Real data: ",data.iloc[36048]['EUI_kBTU_per_sf'])
print("Predict: ",X_test2.iloc[0]['PREDICT'])
print("Mean Absolute Error: ",mean_absolute_error(y_test_new, preds_new))

Real data:  68.2635568063008
Predict:  51.33594512939453
Mean Absolute Error:  9.72474943313263

3D Scatter Plot of Predict datas Vs Real data Vs Error

df3 = y_test_new.to_frame()
df3 = df3.reset_index()

df3new = df3.drop(columns=["index"])
df3new['preds'] = preds_new
df3new['Error'] = abs(df3new['EUI_kBTU_per_sf'] - df3new['preds'])

from mpl_toolkits import mplot3d
import numpy as np
import matplotlib.pyplot as plt


x = df3new['EUI_kBTU_per_sf']
y = df3new['preds']
z = df3new['Error']
 

fig = plt.figure(figsize = (16, 9))
ax = plt.axes(projection ="3d")
   

ax.grid(b = True, color ='grey',
        linestyle ='-.', linewidth = 0.3,
        alpha = 0.2)


my_cmap = plt.get_cmap('inferno_r')
 

sctt = ax.scatter3D(x, y, z,
                    alpha = 0.8,
                    c = z,
                    cmap = my_cmap,
                    marker ='+',s=3)
 
plt.title("3D Scatter Plot", fontdict={'fontsize':18}, pad=12)
ax.set_xlabel('X - axis\n(Real data)', fontweight ='bold')
ax.set_ylabel('Y - axis\n(Predict data)', fontweight ='bold')
ax.set_zlabel('Z - axis\n(Error)', fontweight ='bold')
fig.colorbar(sctt, ax = ax, shrink = 0.5, aspect = 5)

plt.show()

Line Plot of Predict datas, Real data, and Error with 100 observations

plt.figure(figsize=(18,8))
# plt.plot(df3new.iloc[:100])
a = df3new.iloc[:100,0]
b = df3new.iloc[:100,1]
c = df3new.iloc[:100,2]
plt.plot(a, label ="Real data")
plt.plot(b, label ="Predict data")
plt.plot(c, label ="Error")
plt.title("Predict data, Real data", fontdict={'fontsize':18}, pad=12)

plt.legend(loc ="upper right")
plt.show()

import matplotlib.pyplot as plt
plt.figure(figsize=(8, 8))


x = df3new['EUI_kBTU_per_sf']
y = df3new['preds']
y = y.astype("float64")
plt.scatter(x, y, s= 1)

m, b = np.polyfit(x, y, 1)
plt.title("Scatter Plot: Predict data Vs Real data", fontdict={'fontsize':18}, pad=12)

plt.plot(x, m*x + b ,c = "red")
plt.show()